Page transitions: reliability roadmap — re-init registry, upstream the replicas, View Transitions path

[georgeolaru]

Architecture review in plans/2026-06-12-page-transitions-architecture-review.md. Summary:

The Barba-based system is well engineered (head asset diffing, double-init guards, WC exclusions, lifecycle events) but structurally owns the integration burden for the whole ecosystem: every block/plugin needs a special trigger in Anima — cache-busted script re-execution, closure-state vendor hacks (bully), a ~250-line replica of Nova Blocks' header color detection guarded by a MutationObserver, per-plugin allowlists (FacetWP, WC fragments), and resize/scroll replay heuristics. Reliability is capped by the least idempotent script on any page; ecommerce pages are excluded outright.

Roadmap:

• H1 (this codebase, small): invert control — anima.pageTransitions.register({ id, cleanup, reinit }) + anima:before-swap/anima:after-swap events carrying the container; migrate the hardcoded re-init list into registrations; freeze the lifecycle events as public API. Third parties get a documented extension point.
• H2 (Nova Blocks releases): header exposes refresh(container) (delete the replica + observer war), bully gets destroy(), frontend scripts adopt idempotent init(container) — cache-bust reloads become calls.
• H3 (platform): cross-document View Transitions + Speculation Rules as a third engine — real navigations make every plugin/form/analytics/WooCommerce flow work by construction, with the wipe choreography declared in CSS. Ecommerce pages stop being excluded. Prototype first: a CSS-only @view-transition spike answers the choreography-fidelity question in a day. Firefox falls back to plain instant navigation.

Constraint: keep the SM Site Editor Live Site preview working (anima/page_transitions_in_customize_preview, style-manager#139).